# Build a protocol from the study decision

This RWEdnesdays kit contains five editable templates, a model prompt, a review prompt, and a portable skill. The templates are original working aids, not official society forms. Select by what the study is trying to estimate, then check the current official guidance and any sponsor, regulator, ethics, or journal requirements. A dossier organizes evidence for a payer; it does not replace a study protocol.

## Choose the template

| Study goal | Template | Planning guidance | Reporting guidance |
| --- | --- | --- | --- |
| Estimate a comparative treatment effect from existing routine data | Comparative RWE | HARPER and STaRT-RWE; add applicable regulatory protocol guidance | STROBE, RECORD, RECORD-PE |
| Describe frequency, burden, utilization, cost, or treatment patterns | Descriptive RWE | An explicit design/denominator and operational plan; applicable governance | STROBE and RECORD; RECORD-PE for pharmacoepidemiology |
| Synthesize indirect or network treatment evidence | ITC/NMA | PRISMA-P plus ISPOR ITC/NMA conduct and feasibility | PRISMA 2020/NMA and relevance/credibility appraisal |
| Evaluate economic value or payer budget impact | Health economic analysis plan | Economic model good practices and the local decision context; separate BIA module | CHEERS 2022 for economic evaluation; ISPOR BIA for budget impact |
| Assign treatment by randomization in routine care | Pragmatic randomized trial | SPIRIT 2025 and applicable extensions | CONSORT and relevant extensions |

A study spanning goals may need linked protocols or modules. For example, an observational effect estimate can feed an economic model without making their objectives, populations, assumptions, or uncertainty identical. Use separate linked documents when the estimands or governance differ.

## Complete it with or without a model

1. Start with the decision, audience, objective, population, available data, and constraints. Record the evidence cutoff and what results investigators have already seen.
2. Copy the selected template in Word or Markdown. The ZIP includes both formats; the Markdown sources are also the model-ready versions. Fill supported facts; leave unresolved methodological decisions visibly open with an owner. Justify nonapplicable sections rather than silently deleting them.
3. Open the official sources from the template. Keep a versioned crosswalk from each applicable item to a protocol section, evidence artifact, status, and reviewer. Report blocked or unverified sources honestly.
4. Develop the operational definitions, time anchors, analysis, uncertainty, and sensitivity checks together. Link the [data-fitness](https://rwednesdays.com/assets/downloads/rwe-analysis-data-fitness-assessment.md), [phenotype-validation](https://rwednesdays.com/assets/downloads/rwe-phenotype-validation-evidence-matrix.md), and [deviation](https://rwednesdays.com/assets/downloads/rwe-protocol-sap-deviation-log.md) worksheets where relevant.
5. Build an objective-to-analysis-to-output map for the SAP. A generic text summary from Study Designer is an intake brief; expand and review it using the matching template.
6. Obtain human scientific, statistical/economic, data, clinical, and governance review appropriate to the project. Record decisions, versions, and amendments; publication or submission is a separate action.

## Use with a frontier model

For a model that accepts attachments, supply the selected template, these instructions, the approved study brief and data dictionary, the relevant official guidance, and `prompts/build-protocol.md`. Attach only synthetic, public, or otherwise authorized non-sensitive material. Do not upload patient-level data, credentials, or confidential source documents to an unapproved service.

For an agent supporting skills, keep this entire `rwe-protocol-builder` directory together and install it using that agent's documented skill mechanism. `SKILL.md` routes to the local templates and references; no tool-specific orchestration or model-specific hidden behavior is required. The download does not install itself or run scripts. For chat-only models, the copyable prompt is sufficient.

Use `prompts/review-protocol.md` in a fresh conversation with the completed draft and its actual sources. Reviewers should assess unresolved assumptions and whether the design answers the question, not merely whether headings are filled. The skill is a writing aid, not a validated autonomous research system.

## Required handoff

Deliver a protocol draft with version/date and status; an open-decision log; a guidance crosswalk; a claim/source ledger; and the objective-to-analysis-to-output map. Separate verified information, user decisions, modelling assumptions, and unresolved claims. Do not label a protocol final while material design choices or evidence gaps remain unresolved.

## Sources and version control

The references in each template and `references/design-map.md` identify primary guidance checked on 14 October 2026. Verify current status again when preparing an actual protocol. The local checklist mapping does not establish regulatory acceptance. Preserve completed versions and record who approved each consequential change.